Text Filtering And Redaction¶
GuardianNode includes basic text filtering/redaction helpers for some sensitive patterns such as card-like numbers, SSNs, password markers, API-key-like tokens, 2FA-code-like snippets, and private-key blocks.
This is best-effort hygiene, not a privacy certainty.
Parents and administrators should assume captured evidence can contain sensitive on-screen information. Screenshots and extracted text blobs are stored locally and encrypted for parent/admin review, but operational metadata such as app names, URLs, timestamps, risk summaries, categories, and audit details may remain plaintext in SQLite or pending metadata files.
Where It May Run¶
- Backend OCR/event ingestion paths may call
backend/app/services/redaction.py. - Some older code and schemas use the field name
redacted_textfor extracted event text. That name does not mean every sensitive value was removed.
Limits¶
- It does not make secret removal certain for classification.
- It does not make secret removal certain for storage.
- It does not sanitize screenshots.
- It does not replace careful capture-scope and retention settings.
Maintenance¶
Add or adjust patterns in backend/app/services/redaction.py and cover them with
synthetic tests in backend/tests/test_redaction.py. Do not use real child data
or real secrets in tests or issue reports.